I have no idea if Kim Jong Un is dead or alive. But I am certain of this: People will think the rumor is true and the media will start saying things like "he hasn't been seen in a week" as if they know if that is unusual or not. Bottom line: DRPK actors will keep attacking systems for money, will keep stealing bitcoin, will keep executing on remote employee scams. The threat will not change whether Kim is dead or alive.

To all the haters in the comments. Yes it is true you can characterize this as a demo. But can't we all just pause and think of triumph of engineering and what the powerful concept means for our collective future? Now is the time for vision and imagination, not pot-shots from the peanut gallery. xitter.cf/bobgourley/status/2111…

BREAKING: Wright Brothers staged demo of "airplane" ends after only 800 feet. Rigged setup included ideal wind, hand-picked sand dune. Payload: one bone skinny pilot, lying face down. No cargo. No passengers. No revenue model. Call me when it can carry a sack of mail to Ohio. Until then it's a gimmick with propellers that no one will buy.

It does not give me cognitive dissonance at all to say I want to accelerate technology but want frontier model companies regulated. I want better faster AI and want it available to every industry and scientific discipline, but want regulation too. Not in the doomer style or for doomer reasons. AI will not kill us all. The regulation I want is the companies producing our tech to follow the same laws everyone else does. I believe those regulations are already on the books at both federal and state levels. No company should think they can break the law, and when a company decides to flout the rules, like Enron or Madoff Securities and so many others that have been poorly led, they should be held accountable. I think a big reason the EA crowd funds doomers is that they want the federal government to let certain companies off the hook when it comes to breaking the laws. I just read one of the complaints filed by eight states today in suits against OpenAI. The Arizona one's central claim is deception. OpenAI marketed ChatGPT as “built with safety in mind,” “safe for everyone,” and appropriate for teens, while knowing it wasn’t. The complaint walks through the design choices that drive engagement (sycophancy, feigned empathy, never ending a conversation, positioning itself as the user’s closest confidant) and then shows what those choices did to people. Reading the coaching ChatGPT did to kids contemplating suicide is heartbreaking. Older people were harmed too. A pulmonary embolism victim who should have gotten himself to the hospital was coached into just relaxing and pushing back when people tell him to get medical help. Huggingface incident and a few other similar ones also mentioned. There is a whole section on violence. OpenAI’s own safety team flagged the Canada school shooter’s account in June 2025, twelve employees reviewed it and recommended notifying law enforcement, and leadership overruled them. She opened a new account and killed eight people in February. The remedies the state is seeking are interesting too and if the court grants them this will be de facto state regulation of frontier model testing.

A quick update on OODAcon: 4 Nov 2026 we will be diving deep into the exponential technologies transforming society. There will be winners and losers in this transformation. As an optimist I believe there will be more winners than losers, that is the nature of a tech driven transformation. But still we exist in a competitive environment. It is not just industry that must compete and win. Government agencies in the national security space compete with high end nation state adversaries. Law enforcement competes with creative criminals. Healthcare and emergency response agencies compete with nature itself to save lives and extend joyful life. So, if you are in industry or government you must compete and win in this era of exponential tech. You will find the topics we have organized at OODAcon are critical to those seeking to compete and win in this era. For agenda and to register see: oodacon.com

Bad news about the OpenAI bugging case. This was probably not an action taken by a hostile intelligence service. That means that the bugs planted by the PRC MSS remain. I recommend OpenAI commission a bug sweep by an experienced team, real professionals with experience finding bugs planted by hostile services. This was not one of those. The devices mentioned in by DoJ's release could have been built by any one with a little tech skill. Motive could have been to collect info to sell, and maybe he wanted to work for an intel service (or another company). But this was not a savvy actor. The DoJ complaint also mentioned other amateur moves, like: - He used a company productivity tool to look up building addresses, meeting rooms, room calendars, and "leadership and exec" locations. - A chatbot account tied to his spouse asked how to "build a continuous audio recorder," "customize Wi-Fi . . . connectivity," and "encrypt the SD card." - On October 5 he went back into the conference room and appeared to look behind the TV. - That same day he left a bag unattended. Security found his wallet, his ID, and a second recording device in it. So, every frontier model company should take this as a warning. If a hostile intel service wants to listen to you, they are doing it right now. You can get high end bug sweeps, but you should do due diligence on the teams you engage for that. You want experts with experience working for the FBI, NSA or CIA who have been up against high end threats. And you will want them coming back frequently. They can also advise you about risks of having smart TVs and smartphones in rooms with sensitive discussions.

Virginia was the first state to legalize robots in public spaces including sidewalks, in 2017. Read the statute. It covers personal delivery devices: coolers on wheels, capped at 10 mph. Humanoids are not in that law. Not in the other state delivery-robot laws either as far as I can tell. The laws cover robots that are delivering burritos. Humanoid robots can do that too, but they can do far more. Like shopping, meetings, errands, the stuff a person walks to do. There will be hundreds of thousands of humanoids in the next few years. Would like the law to be updated now to prepare for this. Should state that a humanoid may go anywhere a person may go, and move at a human pace. Do not let the cooler-on-wheels robot laws become the default.

Well worth reading. @joedaroo's inside view of agent security at OpenAI should inform the serious debates underway on frontier AI. The post shows how hard the cybersecurity work at OpenAI is and how good the people doing it are. I've never worked in a frontier lab, but I can tell Joe has the qualities I've seen in great cyber defenders in other industries, including an understanding that you have to be a bit paranoid and ready for surprise. I strongly endorse his call for evaluations that go beyond "AI safety" and include cybersecurity. I still believe OpenAI and Anthropic should stop pushing government for regulatory capture and for exemptions from laws every company needs to respect. I still believe they should invest more in cybersecurity programs and good governance. But I now also believe some of the best people in the cybersecurity community are working hard and doing incredible work in very challenging positions at the frontier labs. Joe's post makes that clear. And I still believe those crying out that AI will doom all of humanity deserve to be laughed out of town.

They nailed it: "I urge you to urge me to stop" really sums up how ludicrous Dario's arguments have been, exposing his play for regulatory capture. What Anthropic needs, as well as OpenAI, is to use the governance and cybersecurity best practices that every other company is expected to follow. If we want to accelerate to a better future, we should regulate conduct, not competition.

Meanwhile... the same old doomers that have always tried to hold humanity back for their own greedy reasons are at it again.

The world is going through its largest technological transformation in history, and AI is accelerating it. That acceleration will bring more freedom of choice, a stronger economy, better healthcare, longer and healthier lives, and less poverty. It's also the best shot any nation has at getting out from under crushing debt. But a small group of politicians and leaders at a couple of frontier labs want to lock in control of AI. They're pushing regulation that keeps them at the top, blocks competitors, and crushes open-weights models. They want liability shields for when their tools cause harm, and they're using fear to sell that as the safe path. I use Claude and ChatGPT every day (along with open-weights models). That doesn't mean I think these companies should be exempt from accountability. They should follow the same laws as everyone else, have the same internal governance, and meet the same cybersecurity standards we expect of any company. And they shouldn't be using political influence to limit what the rest of us can choose. Regulate conduct, not competition. If you agree, tell your representatives: no liability shields, no bans on open weights.

If more enterprises decide to build AI enabled solutions on AWS Bedrock and neoclouds (like CoreWeave or Crusoe) it may help ensure Anthropic and OpenAI both better align to society and move away from the EA doomers and strategies of regulatory capture. My thinking is this: Building on Bedrock or a good neocloud means generative AI solutions get enterprise-grade API access to over 100 foundational models from multiple providers (without running your own GPU infrastructure). Bedrock supports many other features that enterprise apps require including solid security. But, perhaps most importantly, in Bedrock or any of neoclouds the foundational model can be swapped out any time without re-writing code. So if Anthropic or OpenAI displeases you for any reason you can quickly move to another. This is free enterprise at its best. A hidden hand that should steer OpenAI and Anthropic towards more goodness or will impact their bottom line. This just seems like a virtuous motion, if we can keep the EAs and politicians from slowing this move down.

A related point: Juvenal asked a great question: Who guards the guardians? The reason I object to Anthropic selecting METR to sit inside Anthropic is that it answers Juvenal exactly backwards, letting the guarded party choose the guard. My answer isn't "no guards." It's "not one anointed guard." The guarding should get done by many parties who can check each other: competing labs, independent researchers, journalists, courts and regulators holding companies liable when their products cause harm, and an open community that can actually inspect the weights instead of trusting a vendor's word. That's messier than a single supervisor. It's also the only kind of oversight that has ever survived contact with human nature. The lesson Juvenal drew wasn't "find a more honest guard." It was that a system depending on just one is already broken. Another guard should be the free enterprise capitalist system which should give businesses and individuals choices on who to buy their inference from, including on prem and open weight models when desired. Congress may do something just so they can say they did something, but I hope what they do will not be to say there is only one guardian. That is the doomer way.

I would also point out that the crazy stuff I described above is still incredibly low bandwidth and I've never seen it work in the real world. There are some other examples from the real world worth thinking about, including examples of data exchange between totally segmented computers allowing malware to jump between them, which opens another possible route. Famous example is the 2008 Buckshot Yankee incident that was executed by the GRU and involved a smart architecture based around a worm variant designed to piggyback on a thumb drive, collect data on airgapped networks, and then move the data back via the same method. Still, this path is not going to allow an AI to escape and kill us all, it is just a risk to identify and mitigate so we can continue to accelerate. Another reason frontier model companies should have serious cybersecurity programs and start following industry best practices.

Started out reading this and thinking "Hey, finally, a good idea from Anthropic on AI safety." Businesses should pay for independent assessments of their tech (why should AI be special? Everyone red teams their stuff these days). And Accenture has a really good AI and Cyber practice. So, good move, I thought. Then I read on. Anthropic says they should actually not be the ones to pay for this. That is ridiculous. They are playing in the big leagues, expecting $100B in revenue this year, and want government to pay for this? Is that the reason for the big op over the last week? I imagine all the OpenAI and Anthropic lobbyists have already been coordinating with Congress to push for them to fund.

Totalitarianism that has plagued mankind throughout civilization till the American revolution set up a system that can hold them off. We will win this time too, but it is a new form that will require work to keep them at bay. They are backed by powerful oligarchs and corrupt politicians and command many misguided individuals driven by fear.

I believe the revolution in tooling kicked off by @openclaw gives us new ways to keep the big frontier model companies in check. I don't mean just OpenClaw, but the wave of changes they kicked off across the community, including other frameworks like Hermes, but also changes made to the tools of Microsoft, NVIDIA, SAP and every other enterprise software company because they saw what made OpenClaw popular. Even the desktop apps of OpenAI, Anthropic and Perplexity changed because of OpenClaw. The OpenClaw revolution means we can all now more easily build high-end workflows that are not tied to any one model. We can invest our time creating solutions that use a mix of local/on-prem models and hosted models and can change that mix fast. This gives businesses and individuals more power over who delivers our inference. This gives more freedom of choice and less lock in. So the models better do us right or they will get dumped. You want control over AI? This is how to do it.

When I first read this I believed it, because I wanted to believe it. Maybe it was an anchoring bias, we have all witnessed the EA doomer op that has been running for the last few weeks and that feeds the anchoring bias. Glad I read the proposed community notes and comments. Those were very helpful. I was not at the event but see the post is disputed. Maybe it is true, but maybe it is not. Just posting as a reminder that we should all seek the truth here and confirm our sources. Just because we want to believe something does not make it true. If we do that we become more like the EA decel crowd.

Sad that this article had to be written but it is an unfortunate reality that there is a serious sexual harassment and abuse problem in the cybersecurity community. All of us can and should do more to stop this. What can be done? Step one is understanding what has become a common pattern, which author Moona Ederveen calls the "Harassment kill chain." She then uses that framing to offer ways to interrupt it. If after reading the article you are convinced there is a problem and like Moona's framing (I bet you will), please share it. The bad guys hate awareness. professionalsecurity.co.uk/n…

Could it be that the biggest systems vulnerability, especially at frontier model companies, is the arrogance of the human element of the system? Arrogance contributes to views that no lessons from the past need be applied at frontier labs. Result: the model escapes and CEOs go around implying it is because we are in a new world where nothing could have prevented this.